CN105631866B - A kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform - Google Patents
A kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform Download PDFInfo
- Publication number
- CN105631866B CN105631866B CN201510989300.1A CN201510989300A CN105631866B CN 105631866 B CN105631866 B CN 105631866B CN 201510989300 A CN201510989300 A CN 201510989300A CN 105631866 B CN105631866 B CN 105631866B
- Authority
- CN
- China
- Prior art keywords
- value
- pixel
- platform
- pixel value
- instruction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform, and heterogeneous platform includes X86-based platform and ARM architecture platform；Subregion is carried out to the pixel of video frame；In first area, adjacent 8 pixels and mixed Gauss model GMM data are imported into register；R, G, B pixel value of each pixel and the difference of mixed Gauss model GMM mean value are calculated separately, above-mentioned squared difference value is calculated, the square value of R, G, B pixel value of 8 pixels is summed respectively；The quadratic sum of R, G, B pixel value is compared with the product of threshold value and the variance of Gaussian Profile, extracts the foreground target in first area；Aforesaid operations are repeated, until the foreground target completed in all areas extracts.The present invention utilizes the multicore of the SSE embedded instructions collection of X86-based platform and the NEON embedded instructions collection of ARM architecture platform and processor；The execution speed for effectively improving algorithm, ensure that the real-time of operation.
Description
Technical field
The present invention relates to video imaging technique fields more particularly to a kind of extraction foreground target method based on heterogeneous platform
Calculation optimization method.
Background technique
Augmented reality (Augmented Reality, AR) is emerging technology developed in recent years, and core is
The content of virtual content and necessary being is subjected to real time fusion, interaction virtual, between reality is formed, to create completely new
Experience.Our company in order to solve the problems, such as online amusement platform main broadcaster because shooting environmental is dull, live video lacks attraction,
A kind of AR method applied to internet video live streaming has been invented, the interest of video can have effectively been promoted, reinforce the vision of video
Impact force.
AR algorithm is applied to PC machine and mobile device such as smart phone, can be significantly in spatially parallel method using hardware
The speed of service of boosting algorithm.Current AR algorithm a large amount of parameter involved in scape target before extraction, passes through C in the prior art
Language calculates formula, is handled point by point image, internal circulating load often, it is computationally intensive, the real-time of operation is very poor.Together
When, existing processing method is identical with the treatment process in mobile device in PC machine, and still, many mobile devices are such as intelligent at present
Mobile phone all uses the processor of ARM framework mostly, PC machine mostly using the processor of X86-based, the instruction set of ARM with
The instruction set of X86 is entirely different；Operation efficiency is reduced using existing processing method.
Summary of the invention
Aiming at the shortcomings existing in the above problems, it is flat based on isomery to provide a kind of extraction foreground target method by the present invention
The calculation optimization method of platform.
To achieve the above object, the present invention provides a kind of extraction calculation optimization of the foreground target method based on heterogeneous platform
Method, comprising:
Step 1: enabling the number of thread process according to the processor nucleus number of the architecture platform type of heterogeneous platform, and select
Instruction fetch collection, the architecture platform type of the heterogeneous platform include X86-based platform or ARM architecture platform；
It include 8 adjacent pixels in each region Step 2: carrying out subregion to the pixel of video frame；
Step 3: adjacent 8 pixels and mixed Gauss model GMM data are imported into register in first area
In, the pixel is directed respectively into from tri- channels R, G, B；
Step 4: calculating separately R, G, B pixel value of each pixel and the difference of mixed Gauss model GMM mean value；
Step 5: the squared difference value is calculated, by the square value of the R pixel value of 8 pixels, square of G pixel value
Value, B pixel value square value sum respectively；
Step 6: be respectively compared the quadratic sum of R, G, B pixel value with the product of threshold value and the variance of Gaussian Profile,
Extract the foreground target in first area；
Step 7: step 3~step 6 is repeated, until the foreground target completed in all areas extracts.
As a further improvement of the present invention, the method for extracting foreground target are as follows:
Each distribution of each pixel of video frame and mixed Gauss model GMM is matched, prospect mesh is extracted
Mark；
The matching formula of pixel and mixed Gauss model GMM are as follows:
|X-μi|≤λσi
Wherein, X is pixel value, μiFor the mean value of Gaussian Profile, σiFor the variance of Gaussian Profile, λ is threshold value；
When pixel and some Gaussian Profile meet above formula, then the pixel matches with Gaussian Profile, is background dot；It is no
It is then foreground point.
As a further improvement of the present invention, the step 1 includes:
In X86-based platform, when processor is monokaryon, single thread processing is enabled；When the processor of X86-based is more
When core, multiple threads are enabled, per thread handles the one part of pixel point of video frame, and chooses SSE embedded instructions collection；
When the processor of ARM architecture platform is monokaryon, single thread processing is enabled；When the processor of ARM framework is multicore
When, multiple threads are enabled, per thread handles the one part of pixel point of video frame, and chooses NEON embedded instructions collection.
As a further improvement of the present invention, in the step 3:
In X86-based platform, with _ mm_loadl_epi64 instruction by adjacent 8 pixels and mixed Gauss model GMM
Data are imported into SSE register；
In ARM architecture platform, adjacent 8 pixels and mixed Gauss model GMM data are imported with vld3_u8 instruction
Into NEON register.
As a further improvement of the present invention, in the step 4:
In X86-based platform, with _ mm_sub_epi16 instruction calculate R, G, B pixel value of each pixel with mix
The difference of Gauss model GMM mean value；
In ARM architecture platform, R, G, B pixel value and mixed Gaussian mould of each pixel are calculated with vabd_u8 instruction
The difference of type GMM mean value.
As a further improvement of the present invention, in the step 5:
In X86-based platform, the squared difference value is calculated with multiplying order _ mm_mullo_epi16, with _ mm_
Adds_epi16 instruction divides the square value of the square value of R pixel value of 8 pixels, the square value of G pixel value, B pixel value
It does not sum；
In ARM architecture platform, the squared difference value is calculated with multiply-add instruction vmlaq_u16, and by 8 pixels
The square value of R pixel value, the square value of G pixel value, B pixel value square value sum respectively.
As a further improvement of the present invention, in the step 6:
In X86-based platform, with compare instruction _ mm_cmpgt_epi16 respectively by the quadratic sum of R, G, B pixel value with
The product λ σ of the variance of threshold value and Gaussian ProfileiIt is compared；
In ARM architecture platform, with compare instruction vcgeq_u16 respectively by the quadratic sum of R, G, B pixel value and threshold value and
The product λ σ of the variance of Gaussian ProfileiIt is compared.
It as a further improvement of the present invention, further include step 8 between the step 3 and step 4；
Step 8: by the data in register from 8 Bits Expandings to 16.
Compared with prior art, the invention has the benefit that
The present invention provides a kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform, before extraction
Scape goal approach can parallel characteristics, it is embedded using the SSE embedded instructions collection of X86-based platform and the NEON of ARM architecture platform
The multicore of instruction set and processor；Wherein SSE embedded instructions collection and NEON embedded instructions collection belong to SIMD instruction collection；Pass through
One instruction handles the mode of multiple data, effectively improves the execution speed of algorithm, reduces the cycle-index of calculating, guarantees
The real-time of operation.
Detailed description of the invention
Fig. 1 is the disclosed foreground target method of extracting of an embodiment of the present invention in the calculation optimization side of X86-based platform
The flow chart of method；
Fig. 2 is the disclosed foreground target method of extracting of an embodiment of the present invention in the calculation optimization side of ARM architecture platform
The flow chart of method.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
The invention discloses a kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform, comprising:
Step 1: enabling the number of thread process according to the processor nucleus number of the architecture platform type of heterogeneous platform, and select
Instruction fetch collection, the architecture platform type of the heterogeneous platform include X86-based platform or ARM architecture platform；
It include 8 adjacent pixels in each region Step 2: carrying out subregion to the pixel of video frame；
Step 3: adjacent 8 pixels and mixed Gauss model GMM data are imported into register in first area
In, the pixel is directed respectively into from tri- channels R, G, B；
Step 4: calculating separately R, G, B pixel value of each pixel and the difference of mixed Gauss model GMM mean value；
Step 5: the squared difference value is calculated, by the square value of the R pixel value of 8 pixels, square of G pixel value
Value, B pixel value square value sum respectively；
Step 6: be respectively compared the quadratic sum of R, G, B pixel value with the product of threshold value and the variance of Gaussian Profile,
Extract the foreground target in first area；
Step 7: step 3~step 6 is repeated, until the foreground target completed in all areas extracts.
Include original background detection, mixed Gauss model foundation, extract foreground target, replacement back in augmented reality AR method
Scape.Wherein: the method for extracting foreground target are as follows: by each point of each pixel of video frame and mixed Gauss model GMM
Cloth is matched, and foreground target is extracted；
The matching formula of pixel and mixed Gauss model GMM are as follows:
|X-μi|≤λσi
X is pixel value, μiFor the mean value of Gaussian Profile, σiFor the variance of Gaussian Profile, λ is threshold value；
When pixel and some Gaussian Profile meet above formula, then the pixel matches with Gaussian Profile, is background dot；It is no
It is then foreground point.
Due to needing to judge whether it is foreground point point by point in the prior art, the step is time-consuming maximum, and the present invention is directed to the step
Suddenly it optimizes.
The present invention is described in further detail with reference to the accompanying drawing:
If input video frame is RGB image, mixed Gauss model is made of k Gaussian Profile, and the value of k is by mobile device
Or the processing capacity of PC machine determines, variance and threshold value are constant.
Embodiment 1: as shown in Figure 1, the present invention, which discloses, extracts foreground target method in the calculation optimization side of X86-based platform
Method, comprising:
S101, when the processor of X86-based be monokaryon when, enable single thread processing；When the processor of X86-based is multicore
When, multiple threads are enabled, per thread handles the one part of pixel point of video frame.
SSE embedded instructions on S102, selection X86-based platform, which integrate, provides instruction as data processing；
X86SSE instruction set is that (Single Instruction, Multiple Data, singly refers to by a kind of SIMD on X86 platform
Enabling, most evidence) instruction set achievees the purpose that raising data-handling efficiency in such a way that an instruction handles multiple data.
S103, subregion is carried out to the pixel of video frame, include 8 adjacent pixels in each region；In the firstth area
In domain, advanced optimized using SSE embedded instructions (intrinsics)；With _ mm_loadl_epi64 instruction by adjacent 8 pictures
The pixel value and mixed Gauss model GMM data of vegetarian refreshments are imported into SSE register；Wherein mixed Gauss model GMM data are
Refer to and is needed when mixed Gauss model GMM is calculated to calculate the parameter of setting and transmitting；Wherein, the pixel value of pixel is from R, G, B
Three channels are directed respectively into SSE register.
S104, with _ mm_cvtepu8_epi16 instruction by the data in SSE register from 8 Bits Expandings to 16, expand to
16 purposes are to generate overflow error in the calculating process prevented below.
S105, the pixel value and mixed Gauss model that tri- channels R, G, B are calculated separately with _ mm_sub_epi16 instruction
The difference of GMM mean value.
S106, the square value that tri- channel difference values of R, G, B are calculated separately with multiplying order _ mm_mullo_epi16, and handle
As a result it is saved in 16 bit registers.
S107, with _ mm_adds_epi16 instruction by the square value of the R pixel value of 8 pixels, square of G pixel value
Value, B pixel value square value sum respectively, by comparing instruction _ mm_cmpgt_epi16 by quadratic sum and constant λ σiCompared
Compared with, extract first area in foreground target；
S108, S101~S107 is repeated, until the foreground target completed in all areas extracts.
Embodiment 2: as shown in Fig. 2, the present invention, which discloses, extracts foreground target method in the calculation optimization side of ARM architecture platform
Method, comprising:
S201, when the processor of ARM framework be monokaryon when, enable single thread processing；When the processor of ARM framework is multicore
When, multiple threads are enabled, per thread handles the one part of pixel point of video frame.
NEON embedded instructions on S202, selection ARM architecture platform, which integrate, provides instruction as data processing；ARM NEON instruction
Collection is that 128 SIMD of one kind on ARM platform (Single Instruction, Multiple Data, single instrction, most evidences) refer to
Collection is enabled, in such a way that an instruction handles multiple data, achievees the purpose that improve data-handling efficiency.
S203, subregion is carried out to the pixel of video frame, include 8 adjacent pixels in each region；In the firstth area
In domain, the pixel value of adjacent 8 pixels and mixed Gauss model GMM data are imported into NEON deposit with vld3_u8 instruction
Device；Wherein the pixel value of pixel is directed respectively into from tri- channels R, G, B into NEON register, can be convenient below to data
It is handled.
After S204, data imported into register, R, G, B pixel value of each pixel are calculated with vabd_u8 instruction and is mixed
Close the difference of Gauss model GMM mean value；
S205, the squared difference value, and putting down the R pixel value of 8 pixels are calculated with multiply-add instruction vmlaq_u16
Side value, the square value of G pixel value, B pixel value square value sum respectively；Again by comparing instruction vcgeq_u16 respectively by R,
G, the product λ σ of the variance of the quadratic sum Yu threshold value and Gaussian Profile of B pixel valueiIt is compared, extracts the prospect in first area
Target；
S206, S201~S205 is repeated, until the foreground target completed in all areas extracts.
The present invention provides a kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform, in X86-based
In platform: in single core processor, enabling single thread, on multi-core processor, enable multithreading；In per thread, use
SSE embedded instructions collection (a kind of SIMD instruction collection), adjacent 8 pixels of a parallel processing.In ARM architecture platform: in list
On core processor, single thread is enabled, on multi-core processor, enables multithreading；In per thread, using NEON embedded instructions
Collect (a kind of SIMD instruction collection), adjacent 8 pixels of a parallel processing.The present invention is according to the extraction foreground target in AR algorithm
Method can parallel characteristics, utilize the SSE embedded instructions collection of X86-based platform and the NEON embedded instructions collection of ARM architecture platform
With the multicore of processor；Wherein SSE embedded instructions collection and NEON embedded instructions collection belong to SIMD instruction collection；Referred to by one
The mode for handling multiple data is enabled, the execution speed of algorithm is effectively improved, reduces the cycle-index of calculating, ensure that operation
Real-time.
These are only the preferred embodiment of the present invention, is not intended to restrict the invention, for those skilled in the art
For member, the invention may be variously modified and varied.All within the spirits and principles of the present invention, it is made it is any modification,
Equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (2)
1. a kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform characterized by comprising
Step 1: enabling the number of thread process, and choose and refer to according to the processor nucleus number of the architecture platform type of heterogeneous platform
Collection is enabled, the architecture platform type of the heterogeneous platform includes X86-based platform or ARM architecture platform；
In X86-based platform, when processor is monokaryon, single thread processing is enabled；When the processor of X86-based is multicore
When, multiple threads are enabled, per thread handles the one part of pixel point of video frame, and chooses SSE embedded instructions collection；
When the processor of ARM architecture platform is monokaryon, single thread processing is enabled；When the processor of ARM framework is multicore, open
With multiple threads, per thread handles the one part of pixel point of video frame, and chooses NEON embedded instructions collection；
SSE embedded instructions collection and NEON embedded instructions collection belong to SIMD instruction collection；
It include 8 adjacent pixels in each region Step 2: carrying out subregion to the pixel of video frame；
Step 3: adjacent 8 pixels and mixed Gauss model GMM data are imported into register, institute in first area
Pixel is stated to be directed respectively into from tri- channels R, G, B；
In the step 3:
In X86-based platform, with _ mm_loadl_epi64 instruction by adjacent 8 pixels and mixed Gauss model GMM data
It imported into SSE register；
In ARM architecture platform, is instructed with vld3_u8 and imported into adjacent 8 pixels and mixed Gauss model GMM data
In NEON register；
Step 4: calculating separately R, G, B pixel value of each pixel and the difference of mixed Gauss model GMM mean value；
In the step 4:
In X86-based platform, R, G, B pixel value and mixed Gaussian of each pixel are calculated with _ mm_sub_epi16 instruction
The difference of model GM M mean value；
In ARM architecture platform, R, G, B pixel value and mixed Gauss model GMM of each pixel are calculated with vabd_u8 instruction
The difference of mean value；
Step 5: the squared difference value is calculated, by the square value of R pixel value, the square value of G pixel value, B of 8 pixels
The square value of pixel value is summed respectively；
In the step 5:
In X86-based platform, the squared difference value is calculated with multiplying order _ mm_mullo_epi16, with _ mm_adds_
Epi16 instruction asks the square value of the square value of R pixel value of 8 pixels, the square value of G pixel value, B pixel value respectively
With；
In ARM architecture platform, the squared difference value is calculated with multiply-add instruction vmlaq_u16, and by the R picture of 8 pixels
The plain square value of value, the square value of G pixel value, B pixel value square value sum respectively；
Step 6: being respectively compared the quadratic sum of R, G, B pixel value with the product of threshold value and the variance of Gaussian Profile, extract
Foreground target in first area；
In the step 6:
In X86-based platform, with compare instruction _ mm_cmpgt_epi16 respectively by the quadratic sum and threshold value of R, G, B pixel value
With the product λ σ of the variance of Gaussian ProfileiIt is compared；
In ARM architecture platform, with compare instruction vcgeq_u16 respectively by the quadratic sum of R, G, B pixel value and threshold value and Gauss
The product λ σ of the variance of distributioniIt is compared；
Step 7: step 3~step 6 is repeated, until the foreground target completed in all areas extracts.
2. extracting calculation optimization method of the foreground target method based on heterogeneous platform as described in claim 1, which is characterized in that
It further include step 8 between the step 3 and step 4；
Step 8: by the data in register from 8 Bits Expandings to 16.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510989300.1A CN105631866B (en) | 2015-12-24 | 2015-12-24 | A kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510989300.1A CN105631866B (en) | 2015-12-24 | 2015-12-24 | A kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105631866A CN105631866A (en) | 2016-06-01 |
| CN105631866B true CN105631866B (en) | 2019-04-05 |
Family
ID=56046752
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510989300.1A Active CN105631866B (en) | 2015-12-24 | 2015-12-24 | A kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105631866B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110428872B (en) * | 2019-09-26 | 2020-03-10 | 深圳华大基因科技服务有限公司 | Method and device for converting gene comparison instruction set |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104572027A (en) * | 2014-12-24 | 2015-04-29 | 天津大学 | Harris feature point detection parallel implementation method based on multi-core processor |
| CN104850866A (en) * | 2015-06-08 | 2015-08-19 | 电子科技大学 | SoC-FPGA-based self-reconstruction K-means cluster technology realization method |
| CN104866295A (en) * | 2014-02-25 | 2015-08-26 | 华为技术有限公司 | Design method and device for OpenCL (open computing language) runtime system framework |
| CN104866537A (en) * | 2014-02-25 | 2015-08-26 | 英特尔公司 | Systems, apparatuses, and methods for feature searching |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8874949B2 (en) * | 2011-12-22 | 2014-10-28 | Intel Corporation | Method, apparatus, and system for energy efficiency and energy conservation including enhanced temperature based voltage control |
-
2015
- 2015-12-24 CN CN201510989300.1A patent/CN105631866B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104866295A (en) * | 2014-02-25 | 2015-08-26 | 华为技术有限公司 | Design method and device for OpenCL (open computing language) runtime system framework |
| CN104866537A (en) * | 2014-02-25 | 2015-08-26 | 英特尔公司 | Systems, apparatuses, and methods for feature searching |
| CN104572027A (en) * | 2014-12-24 | 2015-04-29 | 天津大学 | Harris feature point detection parallel implementation method based on multi-core processor |
| CN104850866A (en) * | 2015-06-08 | 2015-08-19 | 电子科技大学 | SoC-FPGA-based self-reconstruction K-means cluster technology realization method |
Non-Patent Citations (3)
| Title |
|---|
| 基于CUDA加速的运动目标检测;梁良;《中国优秀硕士学位论文全文数据库信息科技辑》;20121015;第2012年卷(第10期);第3.2.1-3.2.3节 |
| 基干AMD APU的终端***的设计与实现;孙文;《中国优秀硕士学位论文全文数据库信息科技辑》;20141130;第2014年卷(第11期);第I137-70页 |
| 龙芯指令***融合技术;胡伟武 等;《中国科学：信息科学》;20150430;第45卷(第4期);第459-479页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105631866A (en) | 2016-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110288614B (en) | Image processing method, device, equipment and storage medium | |
| CN106530265B (en) | A kind of Adaptive image fusion method based on chromaticity coordinate | |
| CN103971330B (en) | Image enchancing method and device | |
| WO2017088365A1 (en) | Skin-colour detection method and apparatus | |
| Kwok et al. | Simultaneous image color correction and enhancement using particle swarm optimization | |
| CN102509099B (en) | Detection method for image salient region | |
| CN104063686B (en) | Crop leaf diseases image interactive diagnostic system and method | |
| CN105513053B (en) | One kind is used for background modeling method in video analysis | |
| CN101859440A (en) | Block-based motion region detection method | |
| CN110827312B (en) | Learning method based on cooperative visual attention neural network | |
| CN107426542A (en) | The hardware structure of Real-Time Arltificial Intelligence video system based on edge calculations | |
| Su et al. | An image enhancement method using the quantum‐behaved particle swarm optimization with an adaptive strategy | |
| CN107465911A (en) | A kind of extraction of depth information method and device | |
| CN107547852A (en) | A kind of big data storage system | |
| CN104574358A (en) | Method and apparatus for scene segmentation from focal stack images | |
| CN113297937A (en) | Image processing method, device, equipment and medium | |
| CN105354228B (en) | Similar diagram searching method and device | |
| CN107506691B (en) | Lip positioning method and system based on skin color detection | |
| He et al. | Integrated image defogging network based on improved atmospheric scattering model and attention feature fusion | |
| CN105631866B (en) | A kind of extraction calculation optimization method of the foreground target method based on heterogeneous platform | |
| Yeganeh et al. | High dynamic range image tone mapping by maximizing a structural fidelity measure | |
| CN111708907B (en) | Target person query method, device, equipment and storage medium | |
| CN114445898B (en) | Face living body detection method, device, equipment, storage medium and program product | |
| CN109741300A (en) | A kind of saliency rapid detection method and device suitable for Video coding | |
| CN105516735A (en) | Representation frame acquisition method and representation frame acquisition apparatus |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |